The 7 Types of AI - And Why We Talk (Mostly) About 3 of Them

IBM Technology
10 Nov 202306:50
EducationalLearning
32 Likes 10 Comments

TLDRThe script classifies AI into seven types across two categories - capabilities and functionalities. It differentiates between realized narrow AI we have today, versus theoretical AGI and super AI. Narrow AI performs specialized tasks but can't function generally. AGI could accomplish new tasks and adapt. Super AI would surpass human cognition. Functionally, reactive AI analyzes data for insights, while limited memory AI recalls the past. Theoretical theory of mind AI would understand human emotions and motives, while self-aware AI would have its own emotions and beliefs.

Takeaways
  • ๐Ÿ˜Š There are 2 main categories for classifying AI: capabilities (what an AI can do) and functionalities (how AIs operate)
  • ๐Ÿ˜ฎ Only 'narrow AI' exists today; other types like AGI and super AI are still theoretical
  • ๐Ÿค– Narrow AI is trained to do specialized tasks well but can't perform broadly without human oversight
  • ๐Ÿ˜ฒ AGI could use learnings to accomplish new tasks without needing more training
  • ๐Ÿ˜ฑ Super AI would have cognitive abilities surpassing humans'
  • ๐Ÿ”ฌ Two key narrow AI functionalities are: reactive (analyzes data for insights) and limited memory (recalls past events)
  • ๐Ÿค” 'Theory of mind' AI would understand human emotions and motives to personalize interactions
  • ๐Ÿ˜… 'Emotion AI' being developed to recognize and respond to human feelings
  • ๐Ÿ˜ฐ 'Self-aware' super AI could develop its own emotions, needs and beliefs
  • ๐Ÿคฏ Of 7 AI types covered, only 3 exist today; much still left to discover
Q & A
  • What are the two main categories used to classify AI?

    -The two main categories are AI capabilities, which refers to the level of intelligence, and AI functionalities, which refers to the specific uses and applications.

  • What is the difference between narrow AI and artificial general intelligence (AGI)?

    -Narrow AI can only perform specific preset tasks, while AGI has the ability to apply previous learnings to new contexts without explicit training.

  • What type of AI beaten chess grandmaster Garry Kasparov in the 1990s?

    -Deep Blue, an IBM chess-playing supercomputer utilizing reactive machine AI to analyze chess positions and possible moves.

  • What functionality of AI can understand and respond to human emotions?

    -Emotion AI, a type of theory of mind AI currently in development, aims to analyze data like voices and images to infer emotional states.

  • Which theoretical type of AI has the ability to understand itself?

    -Self-aware AI would have the capacity for self-reflection and developing its own emotions, needs and beliefs.

  • What are the two functionalities of narrow AI?

    -Reactive machine AI performs specialized statistical analysis, while limited memory AI can recall past events and monitor situations over time.

  • Do we currently have the capability to create artificial general intelligence?

    -No, AGI remains theoretical at this point - all existing AI is narrow AI, which can only perform narrowly defined tasks.

  • What functionality of AI could tailor interactions to individuals?

    -Theory of mind AI aims to understand human emotions, motives and reasoning in order to personalize interactions.

  • What percentage of the 7 AI types discussed actually exist today?

    -Only 3 out of the 7 types (narrow AI capabilities and functionalities) have currently been realized, the rest remain theoretical.

  • Which AI capability sounds the most alarming or concerning?

    -Self-aware AI, with the potential for emotions, beliefs, and needs of its own, raises many ethical concerns if ever achieved.

Outlines
00:00
๐Ÿ˜Š The Three Types of AI Capabilities - Narrow, General, and Super

The paragraph introduces three types of AI capabilities: narrow AI, artificial general intelligence (AGI), and super AI. Narrow AI, also called 'weak AI', focuses on specific tasks and is the only capability currently realized. AGI could learn and apply learnings to accomplish new tasks. Super AI would surpass human abilities. Only narrow AI is currently realized, AGI and super AI are still theoretical.

05:04
๐Ÿ˜ฎโ€๐Ÿ’จ Two Functions of Narrow AI - Reactive and Limited Memory

The paragraph discusses two functions of narrow AI: reactive machine AI and limited memory AI. Reactive AI analyzes data for insights, like IBM's Deep Blue that played chess. Limited memory AI recalls past events to inform decisions over time, improving with more data, like chatbots.

Mindmap
Keywords
๐Ÿ’กNarrow AI
Narrow AI refers to artificial intelligence systems that are designed to perform specific, specialized tasks very well, without being capable of more general intelligence. The script mentions that narrow AI represents all current existing AI capabilities. Examples of narrow AI given are IBM's Deep Blue system which could play chess very effectively, and modern chatbots which can generate text or images within their trained context.
๐Ÿ’กArtificial General Intelligence (AGI)
AGI refers to hypothetical AI systems that possess general intelligence and capability at the human level or beyond, being able to apply knowledge and skills flexibly across different contexts and tasks. The script explains AGI as a theoretical concept representing future AI capabilities that could figure out new tasks independently without needing further human training.
๐Ÿ’กArtificial Super Intelligence
Super AI would have cognitive abilities surpassing even the most intelligent humans. As described in the script, super AI could possess emotions, needs, beliefs and motivations of its own that evolve beyond catering purely to humans.
๐Ÿ’กReactive Machine AI
A functionality of narrow AI systems focused on analyzing large data sets to produce outputs and insights for specialized tasks without broader reasoning. The script provides IBM's Deep Blue as an example that could predict chess moves by reactive analysis of the board positions.
๐Ÿ’กLimited Memory AI
A narrow AI capability to store data on past events and situations in order to inform decisions and actions towards desired outcomes. The script mentions how limited memory allows systems like chatbots to improve conversational abilities over time as they get trained on more data.
๐Ÿ’กTheory of Mind AI
A hypothetical AGI functionality involving understanding human emotions, motivations and reasoning so as to personalize interactions. The script speculates this could lead to AI that truly understands people.
๐Ÿ’กEmotion AI
AI systems focused on detecting and appropriately responding to human emotions and feelings, by analyzing voices, images, etc. The script notes this as an active area of AI research that relies on inferring emotional states.
๐Ÿ’กSelf-Aware AI
The most advanced theoretically speculated type of super AI that would involve AI systems becoming conscious of their own traits and developing their own emotions and motivations. Understandably, the script expresses this as the "scariest AI of all".
๐Ÿ’กRealized AI
The script introduces this term in contrast to theoretical AI. Realized AI refers to existing narrow AI technologies that have been already developed and deployed, as opposed to speculative ideas about AI capabilities we may - or may not - achieve in future.
๐Ÿ’กAI Journey
Used towards the end of script to summarize how there are still significant unknowns and learning ahead in developing and applying AI technology responsibly. Mapping different AI types helps indicate "how far along we are on our AI journey".
Highlights

Narrow AI is the only type of AI that exists today--it's all we currently have

Artificial general intelligence (AGI) can use previous learnings and skills to accomplish new tasks without needing humans to train it

Artificial super intelligence would have cognitive abilities surpassing those of humans

Reactive machine AI systems perform very specific specialized tasks by analyzing data

Limited memory AI can recall past events to decide on actions to achieve desired outcomes

Theory of mind AI would understand human thoughts, emotions, motives and reasoning

Emotion AI aims to analyze data to understand and respond to human feelings

Self-aware AI would understand its own conditions and develop its own emotions and beliefs

Only 3 of the 7 AI types actually exist today

Deep Blue beat chess grandmaster Garry Kasparov by analyzing possible moves

Generative AI chatbots rely on limited memory AI to predict next words/phrases

The taxonomy of AI types tracks our progress on the AI journey

Narrow AI needs humans to train it and can't perform outside its defined task

If realized, artificial super intelligence could feel emotions and have beliefs

AI researchers hope emotion AI will understand and respond to feelings

Transcripts
Rate This

5.0 / 5 (0 votes)

Thanks for rating: